On 1/15/26 6:27 AM, Alex G. wrote: > On Wednesday, January 14, 2026 4:26:36 AM CST Konrad Dybcio wrote: >> On 1/14/26 4:54 AM, Alex G. wrote: >>> On Tuesday, January 13, 2026 8:28:11 AM CST Konrad Dybcio wrote: >>>> On 1/9/26 5:33 AM, Alexandru Gagniuc wrote: >>>>> Support loading remoteproc firmware on IPQ9574 with the qcom_q6v5_wcss >>>>> driver. This firmware is usually used to run ath11k firmware and enable >>>>> wifi with chips such as QCN5024. >>>>> >>>>> When submitting v1, I learned that the firmware can also be loaded by >>>>> the trustzone firmware. Since TZ is not shipped with the kernel, it >>>>> makes sense to have the option of a native init sequence, as not all >>>>> devices come with the latest TZ firmware. >>>>> >>>>> Qualcomm tries to assure us that the TZ firmware will always do the >>>>> right thing (TM), but I am not fully convinced >>>> >>>> Why else do you think it's there in the firmware? :( >>> >>> A more relevant question is, why do some contributors sincerely believe >>> that the TZ initialization of Q6 firmware is not a good idea for their >>> use case? >>> >>> To answer your question, I think the TZ initialization is an afterthought >>> of the SoC design. I think it was only after ther the design stage that >>> it was brought up that a remoteproc on AHB has out-of-band access to >>> system memory, which poses security concerns to some customers. I think >>> authentication was implemented in TZ to address that. I also think that >>> in order to prevent clock glitching from bypassing such verification, >>> they had to move the initialization sequence in TZ as well. >> >> I wouldn't exactly call it an afterthought.. Image authentication (as in, >> verifying the signature of the ELF) has always been part of TZ, because >> doing so in a user-modifiable context would be absolutely nonsensical >> >> qcom_scm_pas_auth_and_reset() which configures and powers up the rproc >> has been there for a really long time too (at least since the 2012 SoCs >> like MSM8974) and I would guesstimate it's been there for a reason - not >> all clocks can or should be accessible from the OS (from a SW standpoint >> it would be convenient to have a separate SECURE_CC block where all the >> clocks we shouldn't care about are moved, but the HW design makes more >> sense as-is, for the most part), plus there is additional access control >> hardware on the platform that must be configured from a secure context >> (by design) which I assume could be part of this sequence, based on >> the specifics of a given SoC > > What was the original use case for the Q6 remoteproc? I see today's use case > is as a conduit for ath11k firmware to control PCIe devices. Was that always > the case? I imagine a more modern design would treat the remoteproc as > untrusted by putting it under a bridge or IOMMU with more strict memory > access > control, so that firmware couldn't access OS memory.
There is an SMMU on this SoC. I don't know the original backstory, but if anything, the through-Q6 approach is probably *more* secure, since there's additional access control hardware inbetween Konrad
